Skip to main content

Monitor Usage, Audit Logs, and Risk

After launch, administrators should regularly review how Siesta AI is being used. Siesta AI provides several operational views for this: Tool Executions, Audit Log, agent Analysis, agent Conversations, agent Feedback, workflow History, and organization token usage.

What to Monitor​

  • User and team changes.
  • New or modified connections.
  • Tool executions, especially write actions.
  • Failed workflow runs.
  • Agents that access sensitive data.
  • Repeated authorization errors.

Tool Executions​

Use Tool Executions when an agent or workflow performs an action through a connected tool. The table shows Agent, Conversation, Action, Status, Resolved at, and Resolved by.

Open the execution detail from the Status column to inspect:

  • arguments passed to the tool,
  • call result,
  • whether approval was required,
  • approval status.

Review statuses such as Pending, Success, Failed, Waiting for approval, and Unknown. For approval workflows, check whether actions are Approved, Rejected, Pending, or marked Approval not required.

Audit Review​

Use audit logs to understand who changed settings, which resources were affected, and when the change happened. Pair Siesta AI audit review with logs from connected systems when investigating external actions.

In Audit Log, use filters for Date Range, User, and Action Type. Open the detail to inspect Record ID, Entity, Date, User, Correlation ID, and the specific changed values.

Agent and Workflow Review​

For production agents, check:

  • Analysis for usage, token consumption, and performance trends.
  • Conversations for real user interactions.
  • Feedback for negative or corrected answers.
  • Evolution before applying prompt changes based on feedback.
  • History to see who changed model, prompt, data, or tool configuration.

For workflows, review History after node changes and verify failed runs in Tool Executions.

Operational Habits​

Review access after onboarding, offboarding, team restructuring, credential rotation, and major workflow changes. For high-impact tools, schedule regular checks with the business owner of the external system.

After rollout changes, review each signal separately: Tool Executions for actions agents performed, Audit Log for configuration and access changes, Feedback for answer quality issues, and Analysis for usage, token, and adoption patterns. Keeping these views separate makes it easier to tell whether a problem came from access, tool behavior, agent quality, or user adoption.